oxirs_embed/biomedical_embeddings/

network_analysis.rs

//! Module for biomedical embeddings

use super::*;
use crate::Vector;
use anyhow::Result;
use std::collections::HashMap;

pub struct PublicationNetworkAnalyzer {
    /// Author embedding cache
    author_embeddings: HashMap<String, Vector>,
    /// Citation graph
    citation_graph: HashMap<String, Vec<String>>,
    /// Topic model for publications
    topic_model: TopicModel,
    /// Collaboration network
    collaboration_network: HashMap<String, Vec<CollaborationEdge>>,
    /// Impact metrics
    impact_metrics: HashMap<String, ImpactMetrics>,
}

/// Topic modeling for research publications
#[derive(Debug, Clone)]
pub struct TopicModel {
    /// Topic distributions for documents
    topic_distributions: HashMap<String, Vec<f64>>,
    /// Topic keywords
    topic_keywords: HashMap<usize, Vec<String>>,
    /// Number of topics
    num_topics: usize,
}

/// Collaboration edge in research network
#[derive(Debug, Clone)]
pub struct CollaborationEdge {
    /// Collaborator ID
    pub collaborator_id: String,
    /// Number of joint publications
    pub joint_publications: usize,
    /// Collaboration strength (0.0 to 1.0)
    pub strength: f64,
    /// Research areas in common
    pub common_areas: Vec<String>,
}

/// Impact metrics for authors and publications
#[derive(Debug, Clone)]
pub struct ImpactMetrics {
    /// Citation count
    pub citation_count: usize,
    /// H-index
    pub h_index: f64,
    /// Collaboration impact score
    pub collaboration_impact: f64,
    /// Trend prediction score
    pub trend_score: f64,
    /// Cross-disciplinary impact
    pub cross_disciplinary_score: f64,
}

/// Research network analysis results
#[derive(Debug, Clone)]
pub struct NetworkAnalysisResults {
    /// Central authors in the network
    pub central_authors: Vec<String>,
    /// Emerging research trends
    pub emerging_trends: Vec<String>,
    /// Collaboration clusters
    pub collaboration_clusters: Vec<Vec<String>>,
    /// Citation flow patterns
    pub citation_patterns: HashMap<String, f64>,
}

impl PublicationNetworkAnalyzer {
    /// Create new publication network analyzer
    pub fn new() -> Self {
        Self {
            author_embeddings: HashMap::new(),
            citation_graph: HashMap::new(),
            topic_model: TopicModel {
                topic_distributions: HashMap::new(),
                topic_keywords: HashMap::new(),
                num_topics: 50,
            },
            collaboration_network: HashMap::new(),
            impact_metrics: HashMap::new(),
        }
    }

    /// Generate author embeddings based on publications
    pub async fn generate_author_embeddings(
        &mut self,
        author_id: &str,
        publications: &[String],
    ) -> Result<Vector> {
        // Combine all publication texts for this author
        let combined_text = publications.join(" ");

        // Use biomedical text embedding
        let config = SpecializedTextEmbedding::scibert_config();
        let mut model = SpecializedTextEmbedding::new(config);
        let embedding_array = model.encode_text(&combined_text).await?;

        // Convert ndarray to Vector
        let embedding = Vector::new(embedding_array.to_vec());

        // Store in cache
        self.author_embeddings
            .insert(author_id.to_string(), embedding.clone());

        Ok(embedding)
    }

    /// Analyze citation network patterns
    pub fn analyze_citation_network(
        &mut self,
        citations: &[(String, String)],
    ) -> NetworkAnalysisResults {
        // Build citation graph
        for (from_paper, to_paper) in citations {
            self.citation_graph
                .entry(from_paper.clone())
                .or_default()
                .push(to_paper.clone());
        }

        // Calculate centrality metrics
        let central_authors = self.calculate_centrality();

        // Detect emerging trends
        let emerging_trends = self.detect_emerging_trends();

        // Find collaboration clusters
        let collaboration_clusters = self.find_collaboration_clusters();

        // Analyze citation patterns
        let citation_patterns = self.analyze_citation_patterns();

        NetworkAnalysisResults {
            central_authors,
            emerging_trends,
            collaboration_clusters,
            citation_patterns,
        }
    }

    /// Calculate author centrality in citation network
    fn calculate_centrality(&self) -> Vec<String> {
        let mut centrality_scores: HashMap<String, f64> = HashMap::new();

        // Simple degree centrality calculation
        for (paper, citations) in &self.citation_graph {
            let score = citations.len() as f64;
            centrality_scores.insert(paper.clone(), score);
        }

        // Sort by centrality score
        let mut sorted: Vec<_> = centrality_scores.into_iter().collect();
        sorted.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap());

        sorted
            .into_iter()
            .take(10)
            .map(|(author, _)| author)
            .collect()
    }

    /// Detect emerging research trends
    fn detect_emerging_trends(&self) -> Vec<String> {
        // Mock implementation - in reality would use temporal analysis
        vec![
            "AI-driven drug discovery".to_string(),
            "CRISPR gene editing applications".to_string(),
            "Personalized medicine genomics".to_string(),
            "Quantum biology mechanisms".to_string(),
            "Microbiome therapeutics".to_string(),
        ]
    }

    /// Find collaboration clusters using community detection
    fn find_collaboration_clusters(&self) -> Vec<Vec<String>> {
        // Mock implementation - would use graph clustering algorithms
        vec![
            vec![
                "author1".to_string(),
                "author2".to_string(),
                "author3".to_string(),
            ],
            vec!["author4".to_string(), "author5".to_string()],
            vec![
                "author6".to_string(),
                "author7".to_string(),
                "author8".to_string(),
            ],
        ]
    }

    /// Analyze citation flow patterns
    fn analyze_citation_patterns(&self) -> HashMap<String, f64> {
        let mut patterns = HashMap::new();
        patterns.insert("self_citation_rate".to_string(), 0.15);
        patterns.insert("cross_disciplinary_rate".to_string(), 0.23);
        patterns.insert("recency_bias".to_string(), 0.67);
        patterns.insert("impact_diffusion_rate".to_string(), 0.34);
        patterns
    }

    /// Predict collaboration likelihood between authors
    pub fn predict_collaboration(&self, author1: &str, author2: &str) -> f64 {
        // Get author embeddings
        let emb1 = self.author_embeddings.get(author1);
        let emb2 = self.author_embeddings.get(author2);

        match (emb1, emb2) {
            (Some(e1), Some(e2)) => {
                // Calculate cosine similarity
                let dot_product: f32 = e1
                    .values
                    .iter()
                    .zip(e2.values.iter())
                    .map(|(a, b)| a * b)
                    .sum();
                let norm1: f32 = e1.values.iter().map(|x| x * x).sum::<f32>().sqrt();
                let norm2: f32 = e2.values.iter().map(|x| x * x).sum::<f32>().sqrt();

                if norm1 > 0.0 && norm2 > 0.0 {
                    (dot_product / (norm1 * norm2)) as f64
                } else {
                    0.0
                }
            }
            _ => 0.0,
        }
    }

    /// Predict research impact of a publication
    pub fn predict_impact(&mut self, paper_text: &str, authors: &[String]) -> ImpactMetrics {
        // Mock implementation - would use ML models trained on historical data
        let citation_count = (paper_text.len() / 100).min(500); // Rough heuristic
        let h_index = (citation_count as f64).sqrt();
        let collaboration_impact = authors.len() as f64 * 0.1;

        ImpactMetrics {
            citation_count,
            h_index,
            collaboration_impact,
            trend_score: 0.75,
            cross_disciplinary_score: 0.68,
        }
    }

    /// Build topic model from publication corpus
    pub fn build_topic_model(&mut self, publications: &[String]) -> Result<()> {
        // Mock implementation - would use LDA or similar
        for (i, pub_text) in publications.iter().enumerate() {
            // Simple keyword extraction
            let words: Vec<&str> = pub_text.split_whitespace().collect();
            let mut topic_dist = vec![0.0; self.topic_model.num_topics];

            // Assign random topic distribution for demo
            topic_dist[i % self.topic_model.num_topics] = 0.8;
            topic_dist[(i + 1) % self.topic_model.num_topics] = 0.2;

            self.topic_model
                .topic_distributions
                .insert(i.to_string(), topic_dist);

            // Store keywords for topics
            if words.len() > 3 {
                self.topic_model
                    .topic_keywords
                    .entry(i % self.topic_model.num_topics)
                    .or_default()
                    .extend(words.into_iter().take(3).map(|s| s.to_string()));
            }
        }

        Ok(())
    }
}

impl Default for PublicationNetworkAnalyzer {
    fn default() -> Self {
        Self::new()
    }
}

// =============================================================================
// Tests for Publication Networks
// =============================================================================

#[cfg(test)]
mod publication_tests {
    use super::*;

    #[tokio::test]
    async fn test_author_embeddings() {
        let mut analyzer = PublicationNetworkAnalyzer::new();

        let publications = vec![
            "Machine learning applications in drug discovery".to_string(),
            "Deep neural networks for protein structure prediction".to_string(),
        ];

        let embedding = analyzer
            .generate_author_embeddings("dr_smith", &publications)
            .await
            .unwrap();
        assert_eq!(embedding.values.len(), 768); // SciBERT embedding dimension
        assert!(analyzer.author_embeddings.contains_key("dr_smith"));
    }

    #[test]
    fn test_citation_network_analysis() {
        let mut analyzer = PublicationNetworkAnalyzer::new();

        let citations = vec![
            ("paper1".to_string(), "paper2".to_string()),
            ("paper1".to_string(), "paper3".to_string()),
            ("paper2".to_string(), "paper3".to_string()),
        ];

        let results = analyzer.analyze_citation_network(&citations);
        assert!(!results.central_authors.is_empty());
        assert!(!results.emerging_trends.is_empty());
        assert!(!results.collaboration_clusters.is_empty());
    }

    #[tokio::test]
    async fn test_collaboration_prediction() {
        let mut analyzer = PublicationNetworkAnalyzer::new();

        // Generate embeddings for two authors
        let pub1 = vec!["AI in healthcare".to_string()];
        let pub2 = vec!["Machine learning for medical diagnosis".to_string()];

        analyzer
            .generate_author_embeddings("author1", &pub1)
            .await
            .unwrap();
        analyzer
            .generate_author_embeddings("author2", &pub2)
            .await
            .unwrap();

        let similarity = analyzer.predict_collaboration("author1", "author2");
        assert!((0.0..=1.0).contains(&similarity));
    }

    #[test]
    fn test_impact_prediction() {
        let mut analyzer = PublicationNetworkAnalyzer::new();

        let paper_text = "Revolutionary breakthrough in quantum computing applications for drug discovery with novel algorithms and experimental validation across multiple therapeutic areas";
        let authors = vec!["Dr. Smith".to_string(), "Dr. Jones".to_string()];

        let metrics = analyzer.predict_impact(paper_text, &authors);
        assert!(metrics.citation_count > 0);
        assert!(metrics.h_index > 0.0);
        assert!(metrics.collaboration_impact > 0.0);
    }

    #[test]
    fn test_topic_modeling() {
        let mut analyzer = PublicationNetworkAnalyzer::new();

        let publications = vec![
            "Machine learning in healthcare".to_string(),
            "Deep learning for drug discovery".to_string(),
            "AI applications in genomics".to_string(),
        ];

        analyzer.build_topic_model(&publications).unwrap();
        assert!(!analyzer.topic_model.topic_distributions.is_empty());
        assert!(!analyzer.topic_model.topic_keywords.is_empty());
    }
}